AutoTruss: Automatic Truss Design with Reinforcement Learning

The official repo of paper 'Automatic Truss Design with Reinforcement Learning'

Generated Truss

The generated trusses are in folder 'assets', you can evaluate them by following command:

python apps/eval.py --config $TRUSS_CONFIG$ --draw-file $TRUSS_PATH$

For example:

python apps/eval.py --config 17_bar_case --draw-file assets/17_bar_1378.txt

This command will generate a truss image.

Run Experiment:

./run.sh without_buckle_case1 test

The command will generate truss automatically and save the truss in 'PostResults/without_buckle_case1/test/'

Or use the following instructions sequentially:

Stage 1

Generating truss layouts in stage 1.

python ./Stage1/continuous_uct_3d.py --config without_buckle_case1 --run-id test

Change output's format to input for stage 2.

python ./Stage1/noise_input_permutation_format_transfer.py --config without_buckle_case1 --run-id test

Stage 2

Use RL to search for lighter truss layouts.

python ./Stage2/main_3d.py --config without_buckle_case1 --run-id test

Add Costomized Config

To generate your own truss with different configs, you can add your own config file in 'configs' and register it in 'config.py'.